Pokee AI has launched PokeeResearch-7B, an open-source 7B deep-research agent designed to reason across the web, retrieve evidence, and self-verify outputs. This innovative tool is now live, complete with code, weights, and a research paper that outlines a reinforcement-learning pipeline utilizing AI feedback and a multi-call reasoning scaffold. The team claims that it achieves state-of-the-art results among its 7B peers on 10 deep-research benchmarks.
Open. Source. SOTA. Deep. Research. 🚀
— Pokee AI (@Pokee_AI) October 22, 2025
Today, we’re releasing PokeeResearch-7B, a SOTA open-source deep research agent that outperforms all other 7B deep research agents. And, we are open-sourcing both the weights and inference code on @huggingface!
We're additionally excited…
Developed for both developers and research teams, the agent can operate locally or via vLLM. It is equipped to integrate web search and content reading, initiate parallel research threads, perform self-checks, and recover from tool errors. The repository includes documentation for A100 80GB testing and offers both CLI and Gradio applications.
Availability begins today. The open-source repository is listed under an Apache-2.0 license, while the research paper mentions MIT, indicating that licensing details may still be in flux. Pokee also provides a preview of a hosted Deep Research API, which is marketed as being more cost-effective than existing providers. The target users include engineers, analysts, and startups that require grounded, citation-rich reports.
Pokee AI positions itself as a foundation-agent company with a focus on reinforcement learning for factual, tool-augmented reasoning. The release of the 7B agent today serves as a public entry point to its agent stack and represents an effort to standardize deep-research workflows across both open and hosted options.
